US EditionOpenAI Has New Agentic Coding Partner for You Now: GPT-5-Codex
GPT-5-Codex dynamically adjusts processing time up to seven hours for coding tasks, improving code reviews with fewer errors and stronger comments, enhancing AI coding tool competitiveness.
- OpenAI announced today the release of GPT-5-Codex, a new coding-optimized variant of GPT-5 integrated into the Codex ecosystem.
- OpenAI developed GPT-5-Codex to improve autonomous coding tasks, addressing developer demand amid a crowded AI coding tools market.
- The model supports long and short coding sessions, runs tests, refactors code in multiple languages, accepts images as input, and integrates deeply with IDEs including VS Code.
- GPT-5-Codex achieved 74.5% accuracy on SWE-bench Verified and scored 51.3% on refactoring tasks, outperforming GPT-5, while usage surged 10 times in one month.
- OpenAI plans an API release, expects Codex to shift developer focus toward oversight and complex decisions, and emphasizes safety through sandboxing and policy controls.
26 Articles
26 Articles
OpenAI foresees millions of AI agents 'somewhere in the cloud' in just a few years — with human supervision
OpenAI launches GPT-5 Codex, enabling AI agents to work on complex software projects.Kim Kyung-Hoon/REUTERSOpenAI foresees millions of AI agents working in companies' data centers under human supervision.OpenAI launches GPT-5 Codex on Monday, which it says can handle complex software projects.OpenAI said human supervision is still needed to keep powerful AI tools aligned with goals.OpenAI is betting the future of software engineering on AI agent…
OpenAI has taken a significant step in developing tools for programmers with the introduction of GPT-5-Codex, an optimized version of its GPT-5 language model focused exclusively on coding tasks. This version is not simply a self-completion tool, but a truly autonomous assistant that includes complex instructions, executes complete development tasks and collaborates in real workflows. The model was trained with data from real programming tasks: …
Coverage Details
Bias Distribution
- 60% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
